Please note that we do not provide an SLA/SLO on indexing. We do not guarantee a build time.
If you are on a shared cluster: the activities of other users can impact your build time. Several users sending peak indexing jobs can slow down all other users. In this case, dedicated infrastructure can improve your build time.
Regardless of whether you are on a shared or dedicated infrastructure, the way the jobs are scheduled and the way the batches are merged affects build time.
- Large indices have an impact on the build time.
- Replicas (real, or even virtual) impact build time: they may need several indices to be built or rebuild, which can lead to a slow down in the indexing process.
- As an example, if we build N indices at most at once, and if you have N replicas, it means that we’ll build an index alone after building the N first indices, leading to double the build time. Thus, the building of other indices can be stopped until the big index is processed. Limiting the number of replicas can help.
- Large batches can also take time to be absorbed. Splitting large batches into small batches can reduce the “stuck time”, but overall will be less efficient.
- More searchable attributes mean more build time
- More attributes for faceting mean more build time